Journal of Criminal Justice 60 (2019) 57–63

Contents lists available at ScienceDirect

Journal of Criminal Justice

journal homepage: www.elsevier.com/locate/jcrimjus

Exploring the impact of 9398 demolitions on neighborhood-level crime in , T ⁎ Matthew Larsona, , Yanqing Xub, Leah Ouelletc, Charles F. Klahm IVd a , Department of Criminal Justice, 3249 Faculty/Administration Building, 656 W. Kirby Street, Detroit, MI 48202, United States b University of Toledo, Department of Geography and Planning, United States c State Appellate Defender Office, United States d Wayne State University, Department of Criminal Justice, United States

ARTICLE INFO ABSTRACT

Keywords: The intersection of neighborhood-level processes and crime has received a wealth of attention in the crimin- Demolitions ological literature over the last century. In line with this tradition, the current study focuses its attention to one Neighborhood Crime of the more recent, and woefully under-explored, policy phenomena embraced by a growing number of cities Crime Policy throughout the United States: demolitions. From 2010 to 2014, the city of Detroit successfully completed a total of 9398 demolitions, making it the nation's leader in the demolitions experiment. Focusing specifically on crime at the block-group level, we examine the association between demolitions and changes in four crime types (i.e. total crime, , drug crime, and ) by calling upon a set of publicly available geo-spatial crime and demolition data. We find that demolitions have a statistically and substantively meaningful negative relationship with total crime, violent crime, and property crime in 2014, net of controls for prior crime and structural covariates. Supplemental analyses also indicate that reductions in crime from 2009 to 2014 were greatest among block-groups that experienced the greatest number of demolitions. We conclude with a dis- cussion of the theoretical and policy implications of demolitions as a potentially valuable crime reduction strategy.

1. Introduction underdeveloped (see Whitaker, 2011). Specifically, there is a significant gap in knowledge as it relates to a byproduct of foreclosure and va- A sizable criminological corpus, dating back to the early 1900s, cancy: property demolition. Although municipalities rely on many in- charges that neighborhood stability influences criminal opportunities novative solutions to deal with distressed properties (Accordino & (Shaw & McKay, 1942). Indeed, seminal works in criminology and Johnson, 2000), several cities, such as, , Cleveland, and Chi- criminal justice highlight the importance of neighborhoods' social and cago have implemented aggressive demolition programs to remove physical environments' roles in explaining crime and criminality vacant and blighted structures. Surprisingly, few empirical studies have (Cohen & Felson, 1979; Jeffery, 1977; Newman, 1972; Shaw & McKay, explored if these campaigns have had any appreciable impact on crime. 1942; Skogan, 1990; Wilson & Kelling, 1982). Even theories that have a The current study explores the impact of a citywide mass demolition geographical orientation, such as social disorganization, primarily focus program on neighborhood-level crime in Detroit, Michigan. Drawing on the social ecology of criminogenic areas. Thus, while Shaw and upon demolition and crime data collected and maintained by the City of McKay (1942) paid limited consideration to neighborhoods' physical Detroit, we assess 1) whether demolitions are associated with crime in features through their descriptive characterizations of the five zones, Detroit at the block-group level and 2) whether change in crime from they ultimately studied the social environments, not physical environ- 2009 to 2014 varies by magnitude of demolition exposure. Our results ments (Jeffery, 1977). The most recent housing crisis in the United suggest that, net of relevant controls of prior crime and structural States has led to a relatively new line of research exploring whether characteristics, demolitions are, on average, associated with lower total home foreclosures and vacancies affect neighborhood-level crime. crime, violent crime, and property crime, but not drug crime. Further, Despite advances in our understanding of the effects of distressed- supplemental analyses show that reductions in crime from 2009 to properties on community-level crime, one related area remains 2014 are greatest among block-groups that experienced the most

⁎ Corresponding author. E-mail address: [email protected] (M. Larson). https://doi.org/10.1016/j.jcrimjus.2018.11.002 Received 6 September 2018; Received in revised form 11 November 2018; Accepted 12 November 2018 Available online 29 November 2018 0047-2352/ © 2018 Elsevier Ltd. All rights reserved. M. Larson et al. Journal of Criminal Justice 60 (2019) 57–63 demolitions. In the end, our findings contribute to the nascent literature the federal Neighborhood Stabilization Program, Spader, Schuetz, and on the effects of demolitions on crime by highlighting the impact of Cortes (2016) found that demolitions reduced theft and in concentrated demolitions within neighborhood block groups. Cleveland, OH (within 250 ft of the demolition), but produced no such effect in Chicago, Illinois. While these early studies report mostly po- 2. Literature review sitive findings, it is difficult to know the true impact of demolitions on crime because there are too few studies available to judge with any Understanding how the social and physical environments of degree of certainty. neighborhoods influence crime and criminality has been the focus of much criminological research. Attention to foreclosure processes 2.1. The case of Detroit emerged in the late 2000s, as the United States faced a housing crisis of epic proportion (Crump et al., 2008), and for good reason: many homes Detroit's landscape has been shaped by waves of abandonment that that enter the foreclosure process become vacant for some period of began in the 1950s, and the consequences of decades' long outward time (Whitaker, 2011). For this reason, we view the foreclosure process migration reached a critical point in 2010. By this time, 22% (79000) of and subsequent vacancies as antecedents to demolitions, and maintain Detroit's structures were abandoned (Mallach, 2012), and the number that each phase of this process has serious implications for neighbor- steadily increased in the aftermath of the nation's foreclosure crisis. By hood well-being. the time Detroit filed for bankruptcy in 2013, there were 85,000 Research has implicated real estate foreclosures as meaningfully abandoned structures (Farley, 2015). The city has long been attempting fi contributing to neighborhood-level crime and disorder. Speci cally, to cope with this abandonment, implementing foreclosure reform (in Teasdale, Clark, and Hinkle (2012) reported subprime lending fore- 1999), launching vacant property auctions and, more recently, fi fi closures were associated with signi cant increases in ve types of adopting a controversial land bank program,1 but it has not staved off crime/disorder in Akron, Ohio. Similarly, Stuckey, Ottensmann, and the blight characterizing many city blocks. Payton (2012) found that foreclosures in Indianapolis, Indiana con- While other rust-belt “legacy” cities have experienced similar pat- tributed to an escalation in crime between 2003 and 2008. Interest- terns of population flight, and felt the pronounced effects of the mort- ingly, research using a longitudinal analysis focused on Glendale, Ar- gage crisis on home values, none were faced with a vacancy crisis of ff izona concluded that the e ect of foreclosures on crime, while Detroit's scale. As a result, Detroit has become leader in removing va- fi fi signi cant, is temporary. Speci cally, the authors reported that in- cant structures quickly and efficiently. In order to achieve a meaningful fi creased foreclosures were associated with signi cant increases in crime, pace of demolition, Detroit reformed its demolition guidelines, but crime rates typically regressed to their normal level within three to streamlining the process and reducing the amount “red-tape” autho- four months depending on crime type (Katz, Wallace, & Hedberg, rities had to navigate (Mallach, 2012). At the start of the program, the 2013). Others have suggested the foreclosure process does not, in and of bulk of the abandoned structures were already foreclosed upon and thus itself, lead to crime. Rather, upticks in crime follow periods of vacancy under the ownership of either the county or the Detroit Land Bank (Cui & Walsh, 2015), or as Immergluck and Smith (2006, p. 854) more Authority, enabling the city to accelerate its demolition process. In “… pointedly lamented, it is through longer-term vacancy and aban- 2014 alone, the first year of Mayor Mike Duggan's, the city demolished ff ” donment that foreclosures a ect neighborhood crime. 3739 structures—239 more than Buffalo, New York's blight removal Indeed, there is ample research suggesting abandoned properties are program demolished in its five-year program (Dynamo Metrics, 2015; hazardous to community-wellbeing. These properties are characterized Yin & Silverman, 2015). Buffalo's demolition process, however, is “ ” as crime attractors (Brantingham & Brantingham, 1995; Shane, 2012) lengthier and requires more resources. Buffalo's demolitions are among “ ” because they serve as criminal hangouts (Boessen & Hipp, 2015; the most expensive in the country, costing the program an average of Sherman, 1993), and invite certain types of crime (e.g., metal scraping, $19,000 per structure (Mallach, 2012). In comparison, Detroit's de- house stripping, urban mining, illegal dumping, squatting, trespassing, molition process costs an average of $12,616 (Kurth, 2017). vandalism; Shane, 2012). In layperson's terms, Holmberg (1998, D1) Many cities, like Detroit, have launched blight removal programs. “ wrote that, crooks, killers and losers tend to infest areas with dead For example, St. Louis, Missouri launched a $13.5 million demolition ” buildings, like maggots on a carcass. Empirically speaking, vacant program in 2017, aiming to demolish 1000 structures within its first fi fl houses have been found to signi cantly in uence burglary, drug year. Similarly, in 2016 Baltimore, Maryland launched Project C.O.R.E. crimes, , vehicle theft, as well as aggravated , , to demolish “as many city blocks of blight as possible,” using $93.5 and homicides (Boessen & Hipp, 2015; Jones & Pridemore, 2016; million local and state funds. Buffalo, New York's “5-in-5” program Raleigh & Galster, 2015). Vacant dwellings not only lead to increased (aiming to demolish 5000 structures in 5 years) from 2007 to 2012 was ff crime, they also a ect property values, as well as municipalities' bud- budgeted for $100 million, $15 million of which were Hardest Hit Fund gets and tax revenue (Joint Center for Housing Studies, 2013). As such, dollars. By comparison, Detroit's program has been awarded $258 many cities that have experienced a disproportionate number of va- million from the Hardest Hit Fund to date (Thibodeau, 2016), making it cancies have implemented demolition plans to address the problem, the largest and most impactful program in the country with over 16,500 often times with aide from the federal government (Hackworth, 2016). structures demolished since 2010. This assessment of Detroit's demoli- In comparison to the foreclosure and vacant properties literature, tion program will be the most comprehensive study to date. the demolition literature is underdeveloped. This is largely a product of the recency of cities' demolition programs. The nascent literate suggests 3. Data & methods demolishing blighted structures produces a crime reduction benefit, although findings vary based on city and unit of analysis. For example, This study relies on crime, demolitions, and neighborhood char- Wheeler, Kim, and Phillips (2018) reported that when measured from acteristics data from Detroit, Michigan to examine the impact of the the microplace (i.e., the demolished structure) demolitions in Buffalo, nation's largest demolitions program on crime at the block-group level. New York produced a significant reduction in part 1 violent and non- In general, it aims to assess whether the over 9000 demolitions com- violent crimes, and the impact was diffused up to 1000 ft from the pleted from 2010 to 2014 were associated with discernible reductions demolition site. However, when they aggregated demolitions to the census tract level, demolitions did not significantly impact crime. Conversely, in Saginaw, Michigan a block group level analysis sug- 1 The Detroit Land Bank Authority is a tax-payer funded office that has been gested demolitions were responsible for a significant reduction in vio- under investigation by various levels of government for “fraud, bid-rigging, lent and property crime (Stacy, 2018). Moreover, in their assessment of environmental violations, and mismanagement” (Neavling, 2017).

58 M. Larson et al. Journal of Criminal Justice 60 (2019) 57–63 in neighborhood crime. More specifically, it focuses on whether de- demographic and socio-economic factors that are established predictors molitions were related with changes in total crime, violent crime, drug of crime. As such, 2010 Census estimates of the following character- crime, and property-based crime in Detroit across that span. The fol- istics of each block-group are included in this study: population density lowing section offers detailed descriptions of the data used to execute (i.e., population/square-mile), median age, percent of households living the current study and the measures that were derived from them. These in poverty, percent female-headed households with children younger data form four specific categories: address-level crime data, address- than 18 years, number of housing units, and number of vacant units. level demolition data, and block-group structural characteristics. 4. Results 3.1. Address-level crime data 4.1. Descriptive statistics The city of Detroit, like many other urban cities throughout the U.S., has increasingly prioritized data transparency in recent years, success- Table 1 reports descriptive statistics for four outcome variables, a fully launching its Open Data Portal in 2015. One body of data included set of three unique demolition variables that capture both the presence in those efforts is crime data from the Detroit Police Department, which and magnitude of demolitions, and a series of six Census variables that includes all reported crime events dating back to January 1st of 2009. effectively measure the structural characteristics of block-groups. In The crime data made available through the portal exist for each re- 2014, Detroit's block-groups experienced an average of 153.74 ported incident and offer a wide-array of information, including, but (SD = 99.77) total crimes, 36.49 (SD = 36.49) violent crimes, 3.68 not limited to, slight offset geographic coordinates (i.e. the address field (SD = 4.08) drug crimes, and 60.98 (SD = 46.69) property crimes. is anonymized by replacing the last two digits of an address with “00”) Table 1 also shows averages for those four crimes in 2009, which were that allow for reported crime incidents to be matched with “place” up to all greater in 2009 than they were in 2014. the block-group level. Given the current study's focus on whether, and Also shown in Table 1, the vast majority of Detroit's block-groups to what extent, demolitions are associated with crime in 2014, net of experienced at least one demolition from 2010 to 2014, with the controls for crime in 2009 and structural factors, ArcGIS 10.6 was used average number of demolitions resting at 10.69 (SD = 14.78). The ex- to match singular crime events to Detroit's 879 block-groups to create perience of demolition, even at this particularly small unit of analysis, annual crime counts across four categories of crime. was much closer to the rule than the exception for neighborhoods Following our main models, which examine the influence of total throughout the city. Interestingly, in one of the few studies on this number of demolitions on four crime types in 2014, we perform sup- subject, Wheeler et al. (2018) modeled the impact of 2035 demolitions plemental analyses to assess change in crime (i.e. 2009 crime counts – on crime in Buffalo, New York from January 2010 to October 2015. 2014 crime counts) across block-groups using a categorical variable Buffalo's exposure to demolitions, while notable for most other cities in that captures magnitude of demolition exposure. The supplemental the U.S., pales in comparison to the over 9000 demolitions that were analyses serve as sensitivity analyses and allow us to contextualize completed in Detroit during a similar, though even shorter, period of whether there is a dose-response relationship between demolitions and time.2 Our five-category variable capturing the magnitude of demoli- crime. Importantly, change in crime from 2009 to 2014 was selected for tion exposure shows that while only 16% of Detroit's block-groups ex- our supplemental analyses because 2009 represents the most proximate perienced no demolitions from 2010 to 2014, 34% experienced one to year-period in which demolitions were not happening with any mean- five demolitions, 15% experienced six to 10 demolitions, 18% experi- ingful regularity. We believe the multi-model approach allows for the enced 11 to 20 demolitions, and 17% experienced at least 21 demoli- most rigorous assessment of the demolitions-crime relationship. tions. Given the linear nature of the four dependent variables and mea- 3.2. Address-level demolition data sures of block-group structural characteristics, a series of t-tests were performed to test for differences across block-groups based on demo- The Detroit Open Data Portal has also prioritized up-to-date detailed lition exposure. These tests revealed both statistically and substantively information on successfully completed demolitions throughout the city. meaningful differences across two of four crime types in 2014, three of It was preceded in those efforts by Data Driven Detroit, which collected four crime types in 2009, and five of six block-group structural char- demolitions data from 2010 to 2013. Using data from these two sources, acteristics (see Table 2). Interestingly, while count of total crime in the current study focuses specifically on demolitions happening from 2009 was statistically greater (p < .05, t-value = −2.45) among 2010 to 2014, the first five years of the city's ramped up efforts to block-groups that experienced demolition from 2010 to 2014 seriously address the blight affecting roughly 80,000 structures. (mean = 209.92, SD = 150.31) than those that did not Notably, the city was not earnestly addressing blight issues via demo- (mean = 176.97, SD = 105.23), such differences were no longer seen lition before 2010. In fact, only 217 demolitions were successfully in 2014 counts of all crime. The same pattern existed for count of completed in 2009, which falls in stark contrast to the nearly 1300 violent crimes by demolition exposure in 2009, when statistical dif- demolitions that completed in 2010 — which, as Fig. 1 highlights, is an ferences existed, and 2014, when no differences were evident. Lastly, annual figure that has only continued to increase. while there was no difference in count of property crime in 2009 based Similar to this study's dependent variables, ArcGIS 10.6 was em- on demolition exposure, statistical differences emerged in 2014 ployed to match demolitions to the block-groups in which they were (p < .05, t-value = 2.16) where counts of property crime became completed. Similar to recent work by Wheeler et al. (2018), we con- lower among block-groups that experienced demolition (mean = 59.47, structed a count of demolitions for all block-groups, which ranged from SD = 48.85) than block-groups that did not (mean = 68.83, a low of 0 to a high of 160. For descriptive and supplemental analyses, SD = 31.58). we developed a five-group categorical variable that includes the fol- Similar to crime, t-tests also revealed sizable differences across lowing demolitions classifications: no demolitions, low demolitions structural characteristics based on demolition exposure. For the 137 (n =1–5), moderate demolitions (n =6–10), high demolitions block-groups that did not experience a single demolition from 2010 to (n =11–20), and very high demolitions (n =21–160) (Fig. 2). 2014, their average population/square-mile, median age, and number of housing units were significantly higher than the 742 block-groups 3.3. Block-group structural characteristics

Given the current study's focus on changes in neighborhood-level 2 Importantly, because there are differences in the populations of Buffalo and crime, it is necessary to include measures of relevant socio- Detroit, these differences exist on a per capita basis as well.

59 M. Larson et al. Journal of Criminal Justice 60 (2019) 57–63

Number of demolitions/year from 2010-2014 4000

3500

3000

2500

2000

1500

1000

500

0 2010 2011 2012 2013 2014

Fig. 1. Total number of demolitions in Detroit per year from 2010 to 2014. that did experience demolition. On the contrary, block-groups that ordinary least squares model after each of our four models to effectively experienced at least one demolition had a larger share of households examine VIF scores. All VIF scores were under 3 and thus well within a living below the poverty line and a greater number of vacant housing healthy range. units than block-groups that had no demolitions. There were no dif- Model 1 in Table 3 specifically describes the relationship between ferences in the percentage of female-headed households with children the number of demolitions between 2010 and 2014 and total crime in under the age of 18 based on demolition exposure. 2014. Net of controls, this model demonstrates that there is a robust relationship between demolitions completed during that span and le- vels of crime across block-groups. Specifically, the Incidence Rate Ratio 4.2. Multivariate analyses for count of demolitions in the total crime model is 0.997 (p < .001, SE = 0.001), meaning that for every demolition, there was a con- A set of four negative binomial models were performed to examine sequent 0.3% reduction in crime. In other words, for roughly every the relationship between the number of demolitions that occurred from three demolitions completed from 2010 to 2014, block-groups experi- 2010 to 2014 and crime in 2014 at the block-group level, net of controls enced an average reduction in crime of almost 1%. Considering that the ff for crime in 2009 (the year preceding Detroit's demolition e orts) and average block-group in this study experienced about 10.7 demolitions fl potential structural in uences. Negative binomial models were the se- during that span, the average reduction in count of all crimes attribu- lected analytic technique because of the over-dispersed nature of our table to demolitions was approximately 3%. Over 35% of block-groups outcome variables. Because negative binomial models in Stata do not experienced more than 10 demolitions, which indicates that those allow for post-estimation assessment of multicollinearity, we ran an

Fig. 2. Block group classifications based on frequency of demolitions.

60 M. Larson et al. Journal of Criminal Justice 60 (2019) 57–63

Table 1 block-groups were likely to have an average reduction in total crime Descriptive statistics across block-groups (N = 879). counts that was even greater.

Mean SD Similar to the total crime model, demolitions were associated with reductions in counts of both violent crime and property crime at the Crime block-group level in 2014. In both cases, the Incident Rate Ratio was Count of all crime (2014) 153.74 99.77 again 0.997, which equates to a one-third of a percent reduction in Count of violent crime (2014) 36.49 19.44 Count of drug crime (2014) 3.68 4.08 violent crime (p < .05, SE 0.001) and property crime (p < .05, Count of property crime (2014) 60.93 46.69 SE = 0.001) for every demolition. While the Incident Rate Ratio of Count of all crime (2009) 204.79 144.66 demolitions is consistent across these two crime types, considering the Count of violent crime (2009) 44.16 23.96 2014 counts of violent crime (mean = 36.49) and property crime Count of drug crime (2009) 5.75 5.93 (mean = 60.93) demonstrates that demolitions from 2010 to 2014 Count of property crime (2009) 86.58 63.79 contributed to a larger reduction in number of property crimes than Demolitions violent crimes. Finally, contrary to 2014 counts of all crime, violent Experienced at least 1 demolition 0.84 – Number of demolitions 10.69 14.78 crime, and property crime, demolitions appear to have no statistically Demolitions-No demolitions (ref) 0.16 – discernable relationship with counts of drug crimes in 2014, although Demolitions-Low (1–5) 0.34 – the measure was approaching significance in the model (Table 4). Demolitions-Moderate (6–10) 0.15 – Demolitions-High (11−20) 0.18 – Demolitions-Very high (21+) 0.17 – 4.3. Supplemental analyses Structural characteristics Population/square-mile in 2010 6542.33 2973.55 A supplementary set of four OLS models were estimated to further Median Age 35.28 6.51 specify the relationship between demolitions and crime in Detroit. Percent below poverty-line 0.41 0.18 Specifically, the supplementary models examine a secondary outcome, Percent female-headed households 0.60 0.33 Number of housing units 397.24 190.66 change in crime from 2009 to 2014, using a categorical measure of de- Number of vacant units 90.70 60.71 molition exposure ranging from 1 (None) to 5 (Very high). Employing an alternative analytic technique and measure of demolitions helps

Table 2 Between-group differences across block-groups by demolition exposure.

Crime No demolitions (n = 137) ⁎ t-value Demolitions (n = 742)

Mean Std. Dev Mean Std. Dev

Count of all crime (2014) 158.06 78.45 0.55 152.94 103.25 Count of violent crime (2014) 34.34 16.69 −1.41 36.89 19.89 Count of drug crime (2014) 2.42 3.57 ⁎⁎ −3.98 3.92 4.13 Count of property crime (2014) 68.83 31.58 ⁎ 2.16 59.47 48.85 Count of all crime (2009) 176.97 105.23 ⁎ −2.45 209.92 150.31 Count of violent crime (2009) 35.56 20.36 ⁎⁎ −4.62 45.75 24.24 Count of drug crime (2009) 3.19 5.51 ⁎⁎ −5.57 6.22 5.88 Count of property crime (2009) 83.02 43.94 −0.71 87.23 66.81

Structural characteristics Population/square-mile in 2010 7289.71 2956.78 ⁎⁎ 3.21 6404.34 2960.03 Median age 38.85 7.42 ⁎⁎ 7.17 34.62 6.11 Percent below poverty-line 0.34 0.19 ⁎⁎ −4.97 0.42 0.17 Percent female-headed households 0.60 0.32 0.24 0.60 0.33 Number of housing units 444.07 297.16 ⁎⁎ 3.14 388.59 162.43 Number of vacant units 69.83 73.18 ⁎⁎ −4.42 94.55 57.34

Between group differences assessed using t-tests. ⁎⁎ p < .01. ⁎ p < .05.

Table 3 Negative binomial models assessing the influence of demolitions on 2014 crime counts.

Total crime (n = 879) Violent crime (n = 879) Drug crime (n = 879) Property crime (n = 879)

Coef IRR SE ⁎ Coef IRR SE ⁎ Coef IRR SE ⁎ Coef IRR SE ⁎

Count of demolitions 0.002 0.997 0.001 ⁎ −0.002 0.997 0.001 ⁎ 0.001 1.000 0.002 −0.003 0.997 0.001 ⁎ Count of 2009 crime (by crime type) 0.002 1.002 0.000 ⁎⁎⁎ 0.011 1.011 0.000 ⁎⁎⁎ 0.057 1.058 0.006 ⁎⁎⁎ 0.005 1.005 0.000 ⁎⁎⁎ Population/square-mile in 2010 0.001 0.999 0.000 ⁎ 0.000 1.000 0.000 0.000 0.999 0.000 ⁎⁎ 0.000 0.999 0.000 Median Age 0.006 0.993 0.002 ⁎ −0.009 0.990 0.002 ⁎⁎⁎ −0.029 0.971 0.001 ⁎⁎⁎ −0.002 0.998 0.002 Percent below poverty-line 0.009 1.009 0.072 0.142 1.152 0.080 −0.006 0.993 0.206 −0.166 0.846 0.076 ⁎ Percent female-headed households 0.087 1.091 0.033 ⁎⁎⁎ 0.138 1.148 0.037 ⁎⁎⁎ 0.138 1.148 0.104 0.088 1.092 0.037 ⁎ Number of housing units 0.001 1.001 0.000 ⁎⁎⁎ 0.001 1.001 0.000 ⁎⁎⁎ 0.000 1.001 0.000 ⁎⁎ 0.001 1.001 0.000 ⁎⁎⁎ Number of vacant units 0.001 0.998 0.000 ⁎⁎⁎ −0.001 0.998 0.000 ⁎⁎⁎ 0.000 1.000 0.000 −0.001 0.998 0.000 ⁎⁎⁎

⁎ p < .05 ⁎⁎ p < .01 ⁎⁎⁎ p < . 001

61 M. Larson et al. Journal of Criminal Justice 60 (2019) 57–63 confirm the stability of the results generated by our main models. 5. Discussion Moreover, they allow us to develop a more specific understand of how the effect of demolitions on crime may vary across the magnitude of The current study explored the impact of the nation's most ag- demolitions that a block-group experienced from 2010 to 2014. gressive demolition program on neighborhood crime in Detroit, Importantly, OLS is employed for the change in crime models because Michigan, adding to the nascent literature on the effects of blight re- the outcome variables are normally distributed, which negative bino- moval policies on neighborhood-level outcomes. Although not without mial models, used for our main models, are not suited for. limitations (discussed below), our main analyses demonstrate that de- A few patterns emerge in the results of the OLS models. First, for molitions from 2010 to 2014 were associated with total crime, violent models predicting change in counts of total crime and property crime crime, and property crime in 2014, net of controls for crime and from 2009 to 2014, all four categories of demolitions were associated structural factors. Our supplemental analyses provide further support with a larger reduction than block-groups that did not experience any for those findings, and also show that block-groups that experienced the demolitions. Specifically, in the total crime model, block-groups that greatest number of demolitions from 2010 to 2014 also witnessed the experienced low, moderate, high, and very high level of demolitions greatest reductions in crime across all four crime types. Concerns over experienced 30.547 (p < .001, SE = 9.123), 27.844 (p < .01, the criminogenic effect of vacant properties have been expressed by SE = 9.075), 23.810 (p < .01, SE = 9.078), and 34.393 (p < .001, scholars for decades, and our study provides additional preliminary SE = 9.963) fewer total crimes in 2014, respectively, relative to block- empirical evidence that removing vacant properties should potentially groups with no demolitions. Further, in the property crime model, be considered as a crime reduction strategy. block-groups that experienced a low, moderate, high, and very high It is important to reiterate that the study location for this project is levels of demolitions demonstrated 12.367 (p < .001, SE = 3.689), unlike any other in the United States in terms of its vacant housing 13.982 (p < .001, SE = 3.508), 11.685 (p < .001, SE = 3.480), and problem and the size and scale of its demolition program. Thus, the 14.717 (p < .001, SE = 3.850) fewer property crimes in 2014, re- results reported here might not be generalizable to other locales. spectively, relative to block-groups where no demolitions occurred. In Although this is a legitimate concern, we maintain the findings are sum, for both types of crime, demolition exposure, regardless of whe- substantively meaningful and contribute to the growing literature in ther that exposure was low or high, contributed to a reduction in crime this area. Again, our main models show that demolitions are associated from 2009 to 2014. Importantly, the significance of these categories with less total crime, violent crime, and property crime, but not drug lends support to the findings demonstrated in our main models. crime. Supplemental provide further support for the relevance of de- Second, a similar pattern of crime reduction benefits was demon- molitions for neighborhood crime, and also show that block-groups that strated in the violent crime model, where three of the four demolitions experienced the greatest number of demolitions were likely to experi- magnitude estimates were significant. While the model showed that a ence the greatest reductions in crime, relative to block-groups with no low level of demolitions from 2010 to 2014 was not related to a re- demolitions, from 2009 to 2014. duction in crime from 2009 to 2014, moderate, high, and very high From a policy perspective, the results suggest that demolition pro- demolitions were associated with reductions of 4.852 (p < .05, grams might be a valid approach for cities facing concentrated vacancy SE = 2.220), 6.148 (p < .01, SE = 2.126), and 6.627 (p < .01, issues. In the case of Detroit, its large-scale program has not only re- SE = 2.325) violent crimes, respectively. Again, these findings support duced the number of vacant properties across the city, it has also led to the results of the negative binomial model in our main analyses that greater reductions in crime in areas where more demolitions have oc- shows an association between demolitions and violent crime in 2014. curred. This suggests, as others have found, that removing distressed Finally, only one of the four demolitions magnitude categories were structures does have an appreciable impact on crime patterns (Stacy, statistically significant in the drug crime model. Specifically, only 2018). Additionally, our results align with literature that implicates block-groups that experienced a very high number (i.e. at least 21) of abandoned properties as “crime attractors” (Brantingham & demolitions saw a reduction in drug crime (b = −1.968, p < .001, SE Brantingham, 1995; Shane, 2012) and “criminal hangouts” (Boessen & =0.621). Again, this finding supports the results of our main models Hipp, 2015; Sherman, 1993). That is, it makes sense that removing that showed no statistically discernible relationship between demoli- places from neighborhoods that generate criminogenic opportunities tions and drug crime in 2014. results in less criminal activity in those areas. Although our results lend to promising conclusions, there are still many unanswered questions regarding the impact of demolitions on crime patterns and the overall wellbeing of neighborhoods. Some of these are a result of limitations

Table 4 OLS models assessing the influence of demolitions on changes in crime counts from 2009 to 2014.

Total crime (n = 879) Violent crime (n = 879) Drug crime (n = 879) Property crime (n = 879)

BSE⁎ BSE⁎ BSE⁎ BSE⁎

Demolitions-No demolitions (ref) –––––––––––– Demolitions-Low (1–5) −30.547 9.123 ⁎⁎⁎ −3.085 1.804 −0.467 0.500 −12.367 3.689 ⁎⁎⁎ Demolitions-Moderate (6–10) −27.844 9.075 ⁎⁎ −4.852 2.220 ⁎ −0.719 0.657 −13.982 3.508 ⁎⁎⁎ Demolitions-High (11–20) −23.819 9.078 ⁎⁎ −6.148 2.126 ⁎⁎ −1.092 0.608 −11.685 3.480 ⁎⁎⁎ Demolitions-Very high (21+) −34.393 9.963 ⁎⁎⁎ −6.627 2.325 ⁎⁎ −1.968 0.621 ⁎⁎⁎ −14.717 3.850 ⁎⁎⁎ Population/square-mile in 2010 0.005 0.001 ⁎⁎⁎ 0.001 0.000 ⁎⁎⁎ 0.000 0.000 0.001 0.000 ⁎⁎ Median Age 1.577 0.573 ⁎⁎ 0.079 0.096 −0.138 0.042 ⁎⁎⁎ 0.875 0.273 ⁎⁎⁎ Percent below poverty-line 14.526 13.968 2.927 3.284 −0.324 1.313 ⁎ 1.355 6.299 Percent female-headed households 18.846 8.780 ⁎ 2.856 1.570 0.352 0.607 9.514 3.452 ⁎⁎ Number of housing units −0.053 0.034 0.007 0.005 0.004 0.002 ⁎ −0.058 0.020 ⁎⁎ Number of vacant units −0.064 0.089 −0.065 0.018 ⁎⁎⁎ −0.022 0.006 ⁎⁎⁎ 0.056 0.046

⁎ p < .05. ⁎⁎ p < .01. ⁎⁎⁎ p < . 001.

62 M. Larson et al. Journal of Criminal Justice 60 (2019) 57–63 associated with the present study, while others are a product of a lack of (2016). Urban blight remediation as a cost-beneficial solution to firearm violence. information about demolition processes, more generally. American Journal of Public Health, 106(12), 2158–2164. Brantingham, P., & Brantingham, P. (1995). Criminality of place. European Journal on With regard to the latter, no studies have explored whether there is Criminal Policy and Research, 3(3), 5–26. a threshold effect in which demolitions beyond a certain point result in Cohen, L. E., & Felson, M. (1979). On estimating the social costs of national economic diminishing returns. Relatedly, there is only one study, to our knowl- policy: A critical examination of the Brenner study. Social Indicators Research, 6(2), fi ff 251–259. edge, that speci es the temporality of the e ects of demolitions on Crump, J., Newman, K., Belsky, E., Ashton, P., Kaplan, D., Hammel, D., & Wyly, E. (2008). crime (see Spader et al., 2016). These are very important details to Cities destroyed (again) for cash: Forum on the U.S. foreclosure crisis. Urban consider for policymakers contemplating demolition programs and ex- Geography, 29(8), 745–784. pecting quick or effective results. It is unlikely that a single demolition Cui, L., & Walsh, R. (2015). Foreclosure, vacancy and crime. Journal of Urban Economics, 87,72–84. of a distressed property within a block group results in meaningful Dixon, J. (2017). After asbestos frustrations, a crackdown on Detroit demolitions. Detroit reductions in crime, but it could be that reductions are visible after Free Press. Available https://www.freep.com/story/news/local/michigan/detroit/ three demolitions, for example. Similarly, there might be an interactive 2017/05/20/detroit-demolitions-blight-asbestos/101520078/. ff Farley, R. (2015). The bankruptcy of Detroit: What role did race play? City and e ect between concentration of demolitions and time so that reductions Community, 14(2), 118–137. in crime are evident only after a certain level of concentration and Frazier, A. E., Bagchi-Sen, S., & Knight, J. (2013). The spatio-temporal impacts of de- – enough time has passed for the neighborhood-level processes to be af- molition land use policy and crime in a shrinking city. Applied Geography, 41,55 64. ff Hackworth, J. (2016). Demolition as urban policy in the American Rust Belt. Environment fected. Other questions remain about whether the e ects are sustained and Planning A: Economy and Space, 48(11), 2201–2222. over time. It is possible that crime reductions are noticeable shortly Immergluck, D., & Smith, G. (2006). The impact of single-family mortgage foreclosures on after completed demolitions but regress to normal levels after a certain neighborhood crime. Housing Studies, 21(6), 851–866. Jeffery, R. (1977). Crime prevention through environmental design. Beverly Hills, CA: SAGE. amount of time. Joint Center for Housing Studies of Harvard University (JCHS) (2013). America ’s Rental Importantly, there are various methodological limitations that the Housing: Evolving Markets and Needs. MA: JCHS: Harvard. current study is constrained by. Perhaps most notably, our analyses are Jones, R. W., & Pridemore, W. A. (2016). A longitudinal study of the impact of home ff vacancy on and burglary rates during the US housing crisis, 2005-2009. cross sectional in nature and cannot bear out any causal e ects. While Crime and Delinquency, 62(9), 1159–1179. this is a serious and legitimate limitation, we view our analyses as ex- Katz, C. M., Wallace, D., & Hedberg, E. C. (2013). A longitudinal assessment of the impact ploratory and representing the first step toward understanding how of foreclosure on neighborhood crime. Journal of Research in Crime and Delinquency, – large-scale demolition programs might affect crime. Also, some might be 50(3), 359 389. Kondo, M. C., Keene, D., Hohl, B. C., MacDonald, J. M., & Branas, C. C. (2015). A dif- concerned that we treat demolitions as an isolated process, even though ference-in-differences study of the effects of a new abandoned building remediation there are undoubtedly potentially relevant events that follow the de- strategy on safety. PLoS One, 10(7), e0129582. fi molition of a home that might be tied to changes in neighborhood-level Kurth, J. (July 6th, 2017). Can Detroit nd salvation through demolition? Bridge Magazine. https://www.crainsdetroit.com/article/20170706/news/633246/can- crime, as well as the tax and mortgage foreclosure processes preceding detroit-find-salvation-through-demolition. home demolition that also affect crime (Immergluck & Smith, 2006; Lacoe, J., & Ellen, I. G. (2015). Mortgage foreclosures and the changing mix of crime in – Lacoe & Ellen, 2015). For instance, a small but growing body of recent micro-neighborhoods. Journal of Research in Crime and Delinquency, 52(5), 717 746. MacDonald, J. (2015). Community design and crime: the impact of housing and the built experimental work suggests that remediation of vacant properties has environment. Crime and justice, 44(1), 333–383. real implications for crime happening in and around those properties, Mallach, A. (2012). Laying the groundwork for change: Demolition, urban strategy, and and they determine these effects even extend to violent crime specifi- policy reform. Brookings Metropolitan Policy Program, 1–47. Metrics, D. (2015). Estimated Home Equity Impacts from Rapid, Targeted Residential cally (Branas et al., 2016; Kondo, Keene, Hohl, MacDonald, & Branas, Demolition in Detroit. MI: Application of a Spatially Dynamic Data System for Decisions 2015). We acknowledge post demolition efforts likely influence col- Support. Detroit, MI: Dynamo Metrics, LLC. lective efficacy, social networking, and opportunities for crime, and Neavling, S. (2017). Detroit Land Bank payroll nearly doubles despite slowdown in de- molitions. Detroit Metro Times. Available. https://www.metrotimes.com/detroit/ that aggressive demolition programs in general may potentially have detroit-land- bank-payroll-nearly-doubles-despite-slowdown-in-demolitions/ other, deleterious impacts on communities. For example, in Detroit Content?oid=4063848. communities have raised concerns over physical health implications Newman, O. (1972). Defensible space. New York, NY: MacMillan. centered on lead and asbestos exposure (Dixon, 2017; MacDonald, Raleigh, E., & Galster, G. (2015). Neighborhood disinvestment, abandonment, and crime dynamics. Journal of Urban Affairs, 37(4), 367–396. 2015). However, trying to account for these efforts in Detroit would be Shane, J. M. (2012). Abandoned buildings and lots. United States Department of Justice. a monumental task. Lastly, although beyond the scope of this paper, we Office of Community Oriented Policing Services. did not address the possibility of crime displacement, which other Shaw, C. R., & McKay, H. D. (1942). Juvenile delinquency and urban areas. Chicago, IL: University of Chicago Press. studies have found (Frazier et al., 2013). Sherman, L. W. (1993). Defiance, deterrence, and irrelevance: A theory of the criminal Undoubtedly, researchers will continue to study the effects of de- sanction. Journal of Research in Crime and Delinquency, 30(4), 445–473. molitions on neighborhood-level crime, disorder, and other social Skogan, W. G. (1990). Disorder and Decline: Crime and the Spiral of Decay in American Cities. New York, NY: The Free Press. processes, especially in light of their continued growth throughout the Spader, J., Schuetz, J., & Cortes, A. (2016). Fewer vacants, fewer crimes? Impacts of country. Future research should attempt to address the aforementioned neighborhood revitalization policies on crime. Regional Science and Urban Economics, limitations. Additionally, scholars should focus e fforts on under- 60,73–84. Stacy, C. P. (2018). The effect of vacant building demolitions on crime under depopu- standing the impact of demolitions on residents' mental and physical lation. Journal of Regional Science, 58(1), 100–115. health, exercise patterns, and overall quality of life. While we have Teasdale, B., Clark, L. M., & Hinkle, J. C. (2012). Subprime lending foreclosures, crime, provided preliminary evidence that concentrated demolitions are as- and neighborhood disorganization: Beyond internal dynamics. American Journal of fi Criminal Justice, 37(2), 163–178. sociated with signi cant reductions in crime, we still need to answer Thibodeau, I. (2016). Detroit razes 10,000th house through demo program. The Detroit whether razing Detroit is actually raising Detroit. News. https://www.detroitnews.com/story/news/local/detroit-city/2016/07/19/ detroit-razes-milestone-house/87286076/. ff References Wheeler, A. P., Kim, D. Y., & Phillips, S. W. (2018). The e ect of housing demolitions on crime in Buffalo, New York. Journal of Research in Crime and Delinquency, 55(3), 390–424. Accordino, J., & Johnson, G. T. (2000). Addressing the vacant and abandoned property Whitaker, S. (2011). Foreclosure-related vacancy rates. Economic Commentary (July). problem. Journal of Urban Affairs, 22(3), 301–315. Wilson, J. Q., & Kelling, G. L. (1982). Broken windows. Atlantic Monthly, 249(3), 29–38. Boessen, A., & Hipp, J. R. (2015). Close-ups and the scale of ecology: Land uses and the Yin, L., & Silverman, R. M. (2015). Housing abandonment and demolition: Exploring the geography of social context and crime. Criminology, 53(3), 399–426. use of micro-level and multi-year models. ISPRS International Journal of Geo- Branas, C. C., Kondo, M. C., Murphy, S. M., South, E. C., Polsky, D., & MacDonald, J. M. Information, 4(3), 1184–1200.

63